Platform vs. open source
Choosing between self-hosting the fishmem engine and FishMem Cloud.
FishMem is open-core. The memory engine is open source and the hosted cloud runs that same engine code — there's no proprietary fork that behaves differently. What you're choosing between is who operates the infrastructure, not which engine you get.
The two options
- Self-host
fishmem— clone the Apache-2.0 engine, wire up your own stores and model providers, and run it on your own infrastructure. See Self-hosting. - FishMem Cloud — the same engine, managed, behind a mem0-compatible API with a dashboard, API keys, metering, and billing.
Side by side
Self-host fishmem | FishMem Cloud | |
|---|---|---|
| Engine / code path | Open-source engine | Same engine, same code path |
| Infrastructure | You operate it (graph store, vector index, model providers) | Fully managed |
| Setup | Configure and run it yourself | Create a key and call the API |
| Dashboard | Not included | Overview, requests, entities, memories, playground |
| API keys & access | You build your own | Project-scoped fm_... keys |
| Usage metering | You instrument it | Built-in credits and usage views |
| Billing | None (your own infra costs) | Stripe subscriptions |
| Scaling & upgrades | Your responsibility | Handled for you |
| License | Apache-2.0 | Commercial service over the same engine |
| Support | Community | Plan-based support |
When to choose which
Self-host when you need data residency or full control, want to run inside your own VPC, are comfortable operating a graph store and a vector index, or are just evaluating the engine locally.
Use the cloud when you'd rather ship product than run infrastructure — you get the dashboard, keys, and metering out of the box, and the engine stays maintained for you.
Because both run the same code, the memory behavior is identical. You can start on one and move to the other; the API surface is the same. The Quickstart gets you running on the cloud in minutes.